Why Global South countries need to care about highly capable AI

Abungu, Cecil; Iradukunda, Marie Victoire; Beggs, Duncan Cass; Hassan, Aquila; Sayidali, Raqda Working Paper Why Global South countries need to care about highly capable AI CIGI Papers, No. 311 Provided in Cooperation with: Centre for International Governance Innovation (CIGI), Waterloo, Ontario Suggested Citation: Abungu, Cecil; Iradukunda, Marie Victoire; Beggs, Duncan Cass; Hassan, Aquila; Sayidali, Raqda (2024) : Why Global South countries need to care about highly capable AI, CIGI Papers, No. 311, Centre for International Governance Innovation (CIGI), Waterloo (Ontario) This Version is available at: https://hdl.handle.net/10419/311790 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ CIGI Papers No. 311 — December 2024 Why Global South Countries Need to Care About Highly Capable AI Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali CIGI Papers No. 311 — December 2024 Why Global South Countries Need to Care About Highly Capable AI Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali About CIGI The Centre for International Governance Innovation (CIGI) is an independent, non-partisan think tank whose peer-reviewed research and trusted analysis influence policy makers to innovate. Our global network of multidisciplinary researchers and strategic partnerships provide policy solutions for the digital era with one goal: to improve people’s lives everywhere. Headquartered in Waterloo, Canada, CIGI has received support from the Government of Canada, the Government of Ontario and founder Jim Balsillie. À propos du CIGI Le Centre pour l’innovation dans la gouvernance internationale (CIGI) est un groupe de réflexion indépendant et non partisan dont les recherches évaluées par des pairs et les analyses fiables incitent les décideurs à innover. Grâce à son réseau mondial de chercheurs pluridisciplinaires et de partenariats stratégiques, le CIGI offre des solutions politiques adaptées à l’ère numérique dans le seul but d’améliorer la vie des gens du monde entier. Le CIGI, dont le siège se trouve à Waterloo, au Canada, bénéficie du soutien du gouvernement du Canada, du gouvernement de l’Ontario et de son fondateur, Jim Balsillie. Credits Executive Director, Global AI Risks Initiative Duncan Cass‑Beggs Senior Research Associate and Program Manager Matthew da Mota Publications Editor Lynn Schellenberg Publications Editor Susan Bubak Graphic Designer Sami Chouhdary Copyright © 2024 by the Centre for International Governance Innovation The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of the Centre for International Governance Innovation or its Board of Directors. For publications enquiries, please contact [email protected]. The text of this work is licensed under CC BY 4.0. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. For reuse or distribution, please include this copyright notice. This work may contain content (including but not limited to graphics, charts and photographs) used or reproduced under licence or with permission from third parties. Permission to reproduce this content must be obtained from third parties directly. Centre for International Governance Innovation and CIGI are registered trademarks. 67 Erb Street West Waterloo, ON, Canada N2L 6C2 www.cigionline.org Table of Contents vi About the Authors vi Acronyms and Abbreviations 1 Executive Summary 1 Introduction 2 The Contours of Highly Capable AI 3 Why Global South Countries Should Care About Highly Capable AI 11 Concluding Recommendations 13 Works Cited vi CIGI Papers No. 311 — December 2024 • Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali About the Authors Cecil Abungu (lead author) is a Ph.D. student at the University of Cambridge and a research affiliate with the Centre for the Study of Existential Risk (University of Cambridge) and the Institute for Law and AI. He is also the coordinator of the ILINA Program. His research interests lie in artificial intelligence (AI) safety governance, algorithmic discrimination, intellectual history, legal theory and constitutional law. Marie Victoire Iradukunda (lead author) is a master of laws student at Harvard Law School and a policy fellow with the Harvard AI Safety Student Team. Her research interests lie in AI governance, particularly the legal duties of AI developers and deployers, and the regulation of AI use by governments in Africa. She has previously served as a research fellow at the ILINA Program. Duncan Cass-Beggs is executive director of the Global AI Risks Initiative at CIGI, focusing on developing innovative governance solutions to address current and future global issues relating to AI. Duncan has more than 25 years of experience working on domestic and international public policy issues, most recently as head of strategic foresight at the Organisation for Economic Co-operation and Development. Aquila Hassan is a project management specialist at the Centre for the Governance of AI in Oxford, England. She has a background in engineering and an interest in research around governing AI for the global majority. Raqda Sayidali is an undergraduate student at Strathmore University and a researcher at the ILINA Program. Her research interests include AI safety governance and the legal challenges surrounding emerging technologies. Acronyms and Abbreviations 4G fourth-generation 5G fifth-generation AGI artificial general intelligence AI artificial intelligence compute computational power ICRC International Committee of the Red Cross IDF Israel Defense Forces LAWS lethal autonomous weapons systems LLMs large language models RLAIF reinforcement learning from AI feedback RLHF reinforcement learning from human feedback UAVs unmanned aerial vehicles 1Why Global South Countries Need to Care About Highly Capable AI Executive Summary By matching and surpassing human cognitive abilities, highly capable artificial intelligence (AI) — advanced AI systems of the foreseeable future, which leading AI companies are working toward as part of their broader goal to create artificial general intelligence (AGI) — could be one of the most transformative technologies the world has ever seen. While this radical technology is being built primarily in Global North countries, its impacts are likely to be felt worldwide, and disproportionately so in those Global South countries with long-standing vulnerabilities — weak state institutions; dependence on labour-intensive, manufacturingbased and export-led economic models; regularly recurring armed conflict; high trust in technology; and more globally subordinated cultures. The authors of this paper consider six ways in which highly capable AI could interact with these vulnerabilities. First, highly capable AI could leave Global South peoples facing a much harsher economic reality. Second, highly capable AI could lead to far more damaging armed conflict in Global South countries. Next, highly capable AI could enable repressive and enduring authoritarianism in Global South countries and could expose Global South peoples to unparalleled manipulation. In addition to that, highly capable AI could also deepen the cultural subordination of Global South peoples. Finally, AI developers and researchers have yet to devise a foolproof way of ensuring that the most advanced models always take actions in line with positive human values. Unless this problem is remedied before the emergence of highly capable AI, there is a chance such AI could lead to catastrophic outcomes including significant loss of life, possibly up to human extinction. Because of the significant societal impacts that highly capable AI could have, being concerned about AI in a general way will not suffice. The authors argue that all stakeholders who care about those who live in Global South countries must pull on the levers available to them with the goal of influencing the ongoing development of highly capable AI. Introduction We live in an era of rapid technological innovation and reach in which the tools and software being built can have a worldwide impact. Historically, milestones such as the telegraph (1837), the telephone (1876) and radio (late 1890s) emerged at intervals of several decades. However, the pace of technological advancement has accelerated significantly. The past 25 years have witnessed a succession of technological leaps (Roser 2023), including fourthand fifth-generation (4G and 5G) mobile networks, increasingly powerful smartphones and the Internet of Things. With few exceptions, the software and devices built as a result of such breakthroughs have found their way to people in every corner of the world. As an example, when Myanmar re-entered global life in 2011 following decades of isolation under military rule, less than one percent of its population had access to the internet. By 2020, this statistic had surged to roughly 44 percent.1 Myanmar essentially experienced a “leapfrog effect,” moving from minimal internet access to widespread smartphone and social media use within a decade. These fast-breaking and far-reaching developments in technology have the capacity to transform societies in unexpected ways, particularly when deployed in diverse cultural contexts. In Myanmar, the rapid spread of the internet had a range of complex consequences, both beneficial and detrimental. On the upside, it provided access to information (Phyo 2023) and new economic opportunities (Jørgensen 2019, 48–50). On the downside, it led to the expansion of platforms that were then used to spread hate speech and disinformation, exacerbating ethnic tensions and violence (Klark and Sagun 2023). AI systems are expected to have an even more profound global impact. Experts anticipate that AI will cause dramatic changes across almost every domain of life, including human rights, knowledge, well-being and governance (Anderson and Rainie 2023). As with previous technological advancements, AI is bound to affect the Global South significantly, especially because the most capable AI systems are being developed in Western countries, where governance frameworks are 1 See Jørgensen (2019) and https://data.worldbank.org/indicator/IT.NET. USER.ZS?locations=MM. 2CIGI Papers No. 311 — December 2024 • Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali shaped and designed with Western contexts and realities in mind (Ayana et al. 2024, 3). There are more than 130 countries in the region known as the “Global South,” each with a diverse society. However, Global South countries2 also share some unique realities, vulnerabilities and challenges. Most of them are faced with similar economic realities, such as severely limited access to capital and investment, high levels of poverty and substantial income inequality (Ayub 2013). Additionally, they share socio-political challenges, including recurring violent conflict (Palik, Obermeier and Rustad 2022), weak state institutions (Choi 2023, 5–7) and lower levels of educated populations.3 Due to the complexity of these issues and the enormous demands of potential solutions, Global South countries have grappled with these problems for decades, and it is reasonable to assume will do so for decades to come. These challenges consume much of the focus and resources of Global South countries, making it difficult for them to address potential risks posed by emerging technologies such as AI. Nonetheless, in this paper, the authors argue that Global South countries must give serious attention to the development of highly capable AI, given the significant context-specific risks it will pose in their societies. This argument is founded on expert predictions regarding the powerful capabilities of highly capable AI systems, and the authors’ analysis of how highly capable AI systems are likely to engage with the ongoing realities and specific vulnerabilities of these countries. In the section following this introduction, “highly capable AI” is defined by canvassing technical predictions about the capabilities that the most powerful AI systems will have in the foreseeable future. The third section then outlines the impacts such AI systems could have in Global South countries. In particular, the paper argues that the development and use of highly capable AI could mean that Global South peoples will face a much harsher economic reality; experience extremely damaging armed conflict; experience repressive and enduring authoritarianism; live through unparalleled manipulation; and find themselves living in a world where their cultures 2 Throughout this paper, the term “Global South countries” will not refer to China, Hong Kong, Macau and Singapore. 3 See Ritchie et al. (2023) and www.worldeconomics.com/Indicator‑Data/ ESG/Social/Mean‑Years‑of‑Schooling/. are very deeply subordinated. The fourth and final section of this paper contains a short discussion on the way forward and forms the conclusion. There is a fair chance that current advanced AI systems could, realistically, give rise to some variation of the risks discussed in this paper. However, this study will show that these risks will be greater and far more concerning with future highly capable AI systems. The authors hope that the research presented here will prompt all stakeholders concerned about Global South peoples to clarify and sharpen their positions on the development of highly capable AI. The Contours of Highly Capable AI In this paper, “highly capable AI” refers to AI systems that demonstrate cognitive capabilities, enabling them to perform economically valuable tasks at or above the level of human beings. Leading AI companies such as Google DeepMind, Meta and OpenAI are working toward building these advanced AI systems as part of their broader goal to create AGI.4 To achieve this goal, they are investing heavily in acquiring data sets, computer chips and data centres (Murgia 2023; Bengio et al. 2024, 843; Gardizy and Efrati 2024). Current advanced AI systems already exhibit impressive capabilities, including what appears to be common-sense reasoning, cause and effect reasoning, step-by-step reasoning and in-context learning (Privitera et al. 2024, 21). They have also demonstrated improved performance on tasks such as image generation and recognition, video generation, and language-based tasks such as text generation, and in fields such as coding, mathematics and biology (Ngo 2023; Bubeck et al. 2023, chapter 2). These impressive capabilities are generally attributed to deep learning (Karnofsky 2016; Piper 2020; Ngo, Chan and Mindermann 2024, 1), and the scaling of model sizes, data sets and computation power (or, compute) used in training (Kaplan et al. 2020, 3; Harris, Harris and 4 See Murgia (2023); Heath (2024); Patel (2023); https://openai.com/ charter. 9Why Global South Countries Need to Care About Highly Capable AI of this paper foresee highly capable AI as likely to be even more able to manipulate users. One key enabler will be highly capable AI’s increased access to context and more personal information of users (Gabriel et al. 2024, 27). Research in persona-based approaches to conversational AI shows that personal information helps the system understand the context of conversation better, making these programs interact more seamlessly and engage in more human dialogue (Liu, Symons and Vatsavai 2022). These capabilities will be scaled up considerably in highly capable AI, making it exceedingly hard for ordinary human users to know whether the AI system they are using is manipulating them. The other enabler will be the abilities that highly capable AI will have to reason, plan and respond to multimodal commands at an inference speed that makes human-machine interactions seem more natural. This would increase the likelihood of anthropomorphism, something that spurs trust and allows manipulation (Gabriel et al. 2024, 96, 102). Taken together, these features will create AI systems that offer an immersive experience and create the illusion of trust by making users feel like they are interacting with a friend or a confidant (ibid., 112). Users in such situations will therefore be more likely to follow any suggestions or directives that the AI system in question proposes. There are already signs that such a world is not far off. Consider, for example, Google’s Gemini,6 which Google says seeks to create a more immersive user experience through agents that can “see and hear what we do, better understand the context we’re in and respond quickly in conversation, making the pace and quality of interactions feel more natural” (cited in Heikkilä 2024). Studies have found that citizens of developing countries are significantly more optimistic and trusting about the impact that AI will have on their lives (Ipsos 2022). In a supplemental report for its 2024 trust barometer, the Edelman Trust Institute (2024, 3–7) reported that trust in technology is higher in developing countries, while it has deteriorated in the United States and the United Kingdom. Another global study found that people in Brazil, China, India and South Africa showed higher levels of trust in AI compared to people in Global North countries (Gillespie 6 Gemini is a “family” of highly capable multimodal models whose quality increases with model size (Gemini Team 2023). et al. 2023, 5). The study particularly noted that people in these four countries trust in the ability, humanity and integrity of AI systems (ibid., 18). The trust AI enjoys in Global South countries seems to be a result of its perceived benefits. Even in cases where participants are fearful about some risks, their excitement, optimism and trust appear to be undiminished (ibid., 24). In India, for instance, technology in general is seen as a solution to many developmental issues and is often trusted as an “authority” in many situations (Kapania et al. 2022). Similarly, in Brazil the level of trust in AI is significantly higher than the level of understanding there is about it (Gillespie et al. 2023, 56). This situation is likely to be replicated in many other Global South countries. Apart from levels of trust in technology, Global South peoples also have lower levels of education in comparison to the rest of the world. For instance, in 2020 the adult literacy rate in Sub-Saharan Africa was only at 66 percent while the global average was 89 percent.7 Studies on misinformation susceptibility have shown that education is a key factor in determining how much someone is susceptible to believing false information. For example, one study identified an inverse statistical correlation between higher levels of education and belief in scientific misinformation (Siani and Green 2023, 8). The same could largely be true for manipulation as well, as education is important in developing the ability to detect and reflect on nuances across judgment domains (Knuutila, Neudert and Howard 2022). Highly Capable AI Could Deepen the Subordination of Global South Peoples’ Cultures Through different projects and approaches, fields such as decolonial studies have shown us that cultural hegemonies can be mapped. In other words, there are identifiable patterns that show the world view dominance of more historically powerful and privileged groups of people over less historically powerful and privileged groups of people (Cortes-Ramirez 2015, 117). This reality has knowingly and unknowingly been brought to bear through coercion or persuasion, often facilitated by technological devices or software. For example, in her book America, as Seen on TV: How Television 7 See https://royalafricansafaris.com/foundation/adult‑education‑in‑ olpalagilagi/. 10 CIGI Papers No. 311 — December 2024 • Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali Shapes Immigrant Expectations Around the Globe, sociologist Clara Rodríguez shows how American television has been able to project US-centric views on social relationships onto the rest of the world (Stoelker 2018). As this paper’s authors argue in the sections that follow, the special capabilities that highly capable AI will possess could lead to a much more deeply entrenched hegemony of Western cultures over Global South peoples’ cultures. Several researchers have already expressed concerns about how current state-of-the-art generative AI models can exacerbate the existing dominance of Western cultures. According to some of this research, the models’ explicit and implicit steering in favour of Western cultures is mostly a result of how they were trained. Consider, for instance, some examples from LLMs, the most cutting-edge AI systems available today. LLMs are trained on data scraped from the internet, which over-represents some parts of the world. Accordingly, the language that is represented enhances cultural alignment where the language in question is prevalent. It is no surprise that some studies have also shown that LLMs’ responses to cognitive psychological tasks most closely resemble those of people from Western, educated, industrialized, rich and democratic societies. Indeed, Rohin Manvi etal. (2024, 1) have recently unveiled research that shows that, due to their training corpora, LLMs are “clearly biased against locations with lower socio-economic conditions (e.g., most of Africa).” Furthermore, the most advanced AI alignment methods — reinforcement learning from human feedback (RLHF) and reinforcement learning from AI feedback (RLAIF) — produce AI systems that are wedded to Western cultures. RLHF involves human judgment in creating preferences in model behaviour, while RLAIF, also called “constitutional AI,” depends on a human-produced set of principles or “constitution” (Conitzer et al. 2024, 1). This means that both methods carry the biases inherent in humans as per their cultural backgrounds (Varshney 2024, 10). For instance, a study examining the cultural biases of LLMs found that the stark differences — a move toward far more secular responses, for example — between the cultural values observable in Open AI’s GPT-3 model and those observable in its immediate successor (GPT-3.5 Turbo) can be attributed to the use of RLHF in training GPT-3.5 Turbo (Tao et al. 2024, 3). Unless there is significant change, the arrival of highly capable AI could worsen the situation for Global South peoples’ cultures. It is likely that just as with existing cutting-edge AI models, highly capable AI will mostly be trained on data sets that have Western cultural biases and norms embedded into them. If RLHF and RLAIF remain the best-performing “alignment” methods available to leading AI developers, we can imagine that both will continue to prop up Western cultures (as explained in the preceding section). And, as many researchers have suggested, highly capable AI could be used everywhere — and to do all sorts of things — in the future. Because of the accuracy and efficiency benefits these AI systems could present, businesses and governments will scramble to integrate them into their work in a way that they have not yet done with existing AI systems. In this event, Global South peoples might be forced to live with AI systems that fundamentally carry Western cultural norms and biases. The cultural biases and norms that will be embedded into highly capable AI systems could be at odds with the cultures of Global South users, leading to misinterpretation and misrepresentation of certain cultures, developments that could in turn create cultural barriers, impose a cultural hegemony and possibly even result in cultural erasure (Prabhakaran, Qadri and Hutchinson 2022, 2). On top of that, by representing certain cultural values as norms through stereotyped responses, AI systems can further entrench these values in users who will then hold them as their own (Anwar et al. 2024, 80–81). Although some may suggest that fine-tuning these AI systems on more localized data could create more “local” models, that would still do little to alter the fundamental philosophical pillars upon which the systems have been built. In other words, if a model has been trained to identify a certain response as wrong or bad, it is exceedingly hard to change that. Others may suggest that the fact that AI could be used to protect languages that are at risk of disappearance shows it might not lead to the subordination of Global South peoples’ cultures (Onome 2024). However, this view conflates the existence of aspects of a culture with its lack of subordination. The two are not necessarily interchangeable. 11Why Global South Countries Need to Care About Highly Capable AI Highly Capable AI Could Inadvertently Result in Mass Death Highly capable AI could also result in catastrophic consequences, such as the death of very many humans (Bengio et al. 2024, 843), and AI misalignment is likely to be the cause. As discussed in the section “The Contours of Highly Capable AI,” highly capable AI systems will be goal-driven. That means that such systems could learn to pursue both desirable and undesirable goals, with the latter being goals not aligned with the widely shared human values (Ngo, Chan and Mindermann 2024, 3). In the main, the pursuit of these undesirable goals will occur where highly capable AI systems have reward functions that are not perfectly aligned with human preferences (ibid.). That possibility is real because the most advanced AI alignment methods still fall short in two ways. First, specifying human intentions using hard-coded reward functions remains very hard and the models may still end up pursuing undesirable goals by exploiting some reward misspecification. Second, although using methods like RLHF might resolve some of these mistakes, feedback from human evaluators is sometimes unreliable, as human evaluators might unintentionally give approval or high reward for undesirable behaviour (ibid.), which could reinforce undesired behaviours in the AI. Similarly, other AI alignment methods being developed by technical researchers still seem a step behind the challenge (Aschenbrenner 2023). Some signs of the risk we face are captured in the small-scale misbehaviour that can be observed in, for example, an AI system that wins a game by exploiting some glitches rather than by playing the game well (Piper 2020; Krakovna et al. 2020). If such a possibility persists, there is a fair chance that the misalignment of highly capable AI systems will result in more large-scale consequences (Ngo, Chan and Mindermann 2024, 8; Ord 2020, 144). These AI systems may pursue instrumental goals including avoiding shutdown, going around human attempts to alter their reward functions, resource acquiring and, eventually, power seeking (Vold and Harris 2021, 735; Ngo, Chan and Mindermann 2024,8; Ord 2020, 145). Ultimately, such AI systems may acquire both the motivation and ability to pursue goals that are incompatible with human well-being such as altering the Earth’s environment to facilitate computing speed or actively disempowering humanity to prevent interference with AI goals. This could eventually lead to catastrophic consequences for humanity, including mass death. Mass casualty events anywhere are inherently bad, but are particularly unjust when caused in Global South countries by technologies developed and primarily benefiting Global North countries. In the worst-case scenario, misaligned highly capable AI systems could cause human extinction, leading not only to the loss of current lives but also of all potential future generations. Such an outcome would be strongly at odds with the value of care for future generations found in the traditions of many ethnic communities in Global South countries.8 Concluding Recommendations The preceding section of this paper laid out the major context-specific risks that Global South peoples could face when highly capable AI is eventually developed and used. Unfortunately, any domestically focused efforts these countries take to counter the risks outlined herein may have limited usefulness. While regulating the use of standalone physical tools and software might be possible, if highly capable AI is in the form of software accessible over the internet (as current LLMs are), any attempts to block access could be circumvented via virtual private networks. As evidence from China shows, even the most sophisticated government-imposed internet firewall can still be ineffective at completely eliminating access to the relevant internet sites (Williams 2023). Global South governments could also attempt to throttle inference but it would be hard for them to know which data centres to target, and the relevant data centres might be outside their jurisdictions (Lehdonvirta, Wú and Hawkins 2024). Global South peoples should be further concerned because their governments are currently illequipped to police the use of AI within their borders. To the authors’ knowledge, at time of writing, no Global South regulatory body has yet conducted an investigation into the use of AI in their society; 8 For one example of communities in southern Africa, see Isaac Schapera (1955, 195–97). 12 CIGI Papers No. 311 — December 2024 • Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali this situation is unlikely to change any time soon because in these countries AI expertise and financial resources are in short supply, while other priorities are endless. This lack of oversight could mean that any interventions might arrive too late in the day. For these reasons, it is critical that Global South countries engage now, while highly capable AI is still being developed. It would make a significant difference if these countries identified and pulled on the levers available to them in order to influence the future development of highly capable AI systems and their potential implications. This could be achieved by, for example: → Appointing expert study groups whose focus is not just AI in general, but highly capable AI specifically. These study groups could give Global South countries advice on what specific steps to take with regard to highly capable AI. → Ensuring that their national and regional policies (and eventually laws) pay special attention to preventing and mitigating the risks that highly capable AI could pose to their societies. → Forming coalitions of like-minded nations and pushing the leading AI companies and Western countries to pay special attention to the global scale of impacts of highly capable AI as well as the context-specific implications for Global South countries. This work could be done through international fora or state-to-state diplomacy. Outside of government, researchers and civil society in Global South countries can make a significant difference if they begin engaging as well. Their voices and influence could pressure their governments, pressure Western governments, or persuade researchers and civil society organizations in Western countries to press their governments to ask the leading AI companies more questions. Those who care about the fate of Global South peoples must realize that getting serious about AI generally is not enough. It is crucial that we specifically get serious about the risks of highly capable AI. Acknowledgements The authors would like to specially thank Cullen O’Keefe, Faith Gakii, Lynn Schellenberg, Matthew da Mota, Seán Ó hÉigeartaigh, Sienka Dounia, Sumaya Nur and Vicky Miriti for their feedback and assistance in the process of writing this paper. 13Why Global South Countries Need to Care About Highly Capable AI Works Cited Allen, Greg and Taniel Chan. 2017. “Artificial Intelligence and National Security.” July. Belfer Center Study. Cambridge, MA: Belfer Center for Science and International Affairs, Harvard Kennedy School. www.belfercenter.org/ publication/artificial-intelligence-and-national-security. Allen, Karen. 2021. “Drones and Violent Nonstate Actors in Africa.” Africa Center for Strategic Studies, August 6. https://africacenter.org/spotlight/ drones-and-violent-nonstate-actors-in-africa/. Allen, Nathaniel and Marian “Ify” Okpali. 2022. “Artificial intelligence creeps on to the African battlefield.” Brookings Institution, February 2. www.brookings.edu/articles/ artificial-intelligence-creeps-on-to-the-african-battlefield/. Amnesty International. 2024. The State of the World’s Human Rights: April 2024. London, UK: Amnesty International. www.amnesty.org/en/documents/pol10/7200/2024/en. Anderson, Janna and Lee Rainie. 2023. As AI Spreads, Experts Predict the Best and Worst Changes in Digital Life by 2035. Pew Research Center, June 21. www.pewresearch.org/ internet/2023/06/21/as-ai-spreads-experts-predictthe-best-and-worst-changes-in-digital-life-by-2035/. Anwar, Usman, Abulhair Saparov, Javier Rando, Daniel Paleka, Miles Turpin, Peter Hase, Ekdeep Singh Lubana et al. 2024. “Foundational Challenges in Assuring Alignment and Safety of Large Language Models.” Preprint, arXiv, September6. https://doi.org/10.48550/arXiv.2404.09932. Armstrong, Adrienne. 1981. “The Political Consequences of Economic Dependence.” Journal of Conflict Resolution 25 (3): 401–28. www.jstor.org/stable/173661. Arriola, Leonardo R., Lise Rakner and Nicolas van de Walle. 2023. Democratic Backsliding in Africa? Autocratization, Resilience, and Contention. Oxford, UK: Oxford University Press. Aschenbrenner, Leopold. 2023. “Nobody’s on the ball on AGI alignment.” For Our Prosperity (blog), March 29. www.forourposterity.com/nobodys-onthe-ball-on-agi-alignment/. Ayana, Gelan, Kokeb Dese, Hundessa Daba Nemomssa, Bontu Habtamu, Bruce Mellado, Kingsley Badu, Edmund Yamba et al. 2024. “Decolonizing global AI governance: assessment of the state of decolonized AI governance in Sub-Saharan Africa.” Royal Society Open Science 11(8): 231994. https://doi.org/10.1098/rsos.231994. Ayres, Ian and Jack M. Balkin. 2024. “The Law of AI is the Law of Risky Agents without Intentions.” Preprint, Social Science Research Network, June 12. https://papers.ssrn.com/ sol3/papers.cfm?abstract_id=4862025. Ayub, Mahmood. 2013. “Poverty and Inequality.” Global Journal of Emerging Market Economies 5 (3): 329–46. https://doi.org/10.1177/0974910113505796. Beaumont, Hilary. 2022. “‘Never sleeps, never even blinks’: the hi-tech Anduril towers spreading along the US borders.” The Guardian, September 16. www.theguardian.com/ us-news/2022/sep/16/anduril-towerssurveillance-us-mexico-border-migrants. Bell, Stephanie A. and Anton Korinek. 2023. “AI’s Economic Peril.” Journal of Democracy 34 (4): 151–61. www.journalofdemocracy.org/articles/ais-economic-peril/. Bengio, Yoshua. 2023. “AI and Catastrophic Risk.” Journal of Democracy 34 (4):111–21. www.journalofdemocracy.org/ articles/ai-and-catastrophic-risk/. Bengio, Yoshua, Geoffrey E. Hinton, Andrew Yao, Dawn Song, Pieter Abbeel, Trevor Darrell, Yuval N. Harari et al. 2024. “Managing extreme AI risks amid rapid progress.” Science 384 (6698): 842–45. https://doi.org/10.1126/science.adn0117. Branwen, Gwern. 2022. “The Scaling Hypothesis.” Gwern.net, January 2. https://gwern.net/scaling-hypothesis. Brundage, Miles, Shahar Avin, Jack Clark, Helen Toner, Peter Eckersley, Ben Garfinkel, Allan Dafoe et al. 2018. The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation. Preprint, arXiv, February 20. https://doi.org/10.48550/arXiv.1802.07228. Bubeck, Sébastien, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee et al. 2023. “Sparks of Artificial General Intelligence: Early experiments with GPT-4.” Preprint, arXiv, April 13. https://arxiv.org/abs/2303.12712. Campbell-James, Jonathan. 2024. “Curbing Outside Intervention in the Sudan War.” Policy analysis, April 4. PolicyWatch 3855. Washington, DC: The Washington Institute for Near East Policy. www.washingtoninstitute.org/policyanalysis/curbing-outside-intervention-sudan-war. Caparini, Marina and Ian Davis. 2022. “Armed conflict and peace processes in the Americas.” In SIPRI Yearbook 2022: Armaments, Disarmament and International Security, edited by Stockholm International Peace Research Institute, 75–102. Oxford, UK: Oxford University Press. www.sipri.org/yearbook/2022/03. 14 CIGI Papers No. 311 — December 2024 • Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali Carlsmith, Joseph. 2024. “Is Power-Seeking AI an Existential Risk?” Preprint, arXiv, August 13. https://doi.org/10.48550/ arXiv.2206.13353. Carroll, Micah, Alan Chan, Henry Ashton and David Krueger. 2023. “Characterizing Manipulation from AI Systems.” Preprint, arXiv, October 30. https://doi.org/10.48550/arXiv.2303.09387. Casals, Marcelo. 2024. “Cold War Dictatorships in the Southern Cone (Brazil, Argentina, Uruguay, Chile).” Oxford Bibliographies, June 21. www.oxfordbibliographies.com/ display/document/obo-9780199791279/ obo-9780199791279-0254.xml. Cazzaniga, Maura, Florence Jaumotte, Longji Li, Giovanni Melina, Augustus J. Panton, Carlo Pizzinelli, Emma Rockall and Marina M. Tavares. 2024. “Gen-AI: Artificial Intelligence and the Future of Work.” IMF Staff Discussion Note SDN/2024/001. Washington, DC: International Monetary Fund. www.imf.org/en/Publications/ Staff-Discussion-Notes/Issues/2024/01/14/Gen-AIArtificial-Intelligence-and-the-Future-of-Work-542379. Center for Preventive Action. 2024. “Conflict in the Democratic Republic of Congo.” Council on Foreign Relations, June 20. www.cfr.org/global-conflict-tracker/ conflict/violence-democratic-republic-congo. Chan, Alan, Rebecca Salganik, Alva Markelius, Chris Pang, Nitarshan Rajkumar, Dmitrii Krasheninnikov, Lauro Langosco et al. 2023. “Harms from Increasingly Agentic Algorithmic Systems.” Preprint, arXiv, May 12. https://doi.org/10.48550/arXiv.2302.10329. Choi, Anthony. 2023. “The Global North and Global South in Political Dilemma: Postmaterialist Effects of Age on Government Trust.” Gettysburg, PA: The Cupola: Scholarship at Gettysburg College. https://cupola.gettysburg.edu/student_scholarship/1113. Collier, Ben. 2024. “Targeted social media ads are influencing our behaviour — and the government uses them too.” The Conversation, February 27. https://theconversation.com/ targeted-social-media-ads-are-influencing-our-behaviourand-the-government-uses-them-too-223576. Collier, Ruth B. 1982. Regimes in Tropical Africa: Changing Forms of Supremacy, 1945–1975. Berkley, CA: University of California Press, 1982. Conitzer, Vincent, Rachel Freedman, Jobst Heitzig, Wesley H. Holliday, Bob M. Jacobs, Nathan Lambert, Milan Mossé et al. 2024. “Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback.” Preprint, arXiv, June 4. https://doi.org/10.48550/arXiv.2404.10271. Cortes-Ramirez, Eugenio-Enrique. 2015. “Cultural Hegemony Today. From Cultural Studies to Critical Pedagogy.” Postcolonial Directions in Education 4 (2): 116–39. https://core.ac.uk/download/pdf/84894568.pdf. Cremer, Carla Zoe and Jess Whittlestone. 2021. “Artificial Canaries: Early Warning Signs for Anticipatory and Democratic Governance of AI.” International Journal of Interactive Multimedia and Artificial Intelligence 6 (5): 100–109. https://dx.doi.org/10.9781/ijimai.2021.02.011. Criddle, Cristina, Madhumita Murgia and George Hammond. 2024. “Artificial intelligence breakthroughs create new ‘brain’ for advanced robots.” Financial Times, July 26. www.ft.com/ content/bea9df71-371c-4045-9cb4-64c22789bf7b. Cummings, M. L., Heather M. Roff, Kenneth Cukier, Jacob Parakilas and Hannah Bryce. 2018. Artificial Intelligence and International Affairs: Disruption Anticipated. Chatham House Report, June. www.chathamhouse.org/sites/default/files/ publications/research/2018-06-14-artificial-intelligenceinternational-affairs-cummings-roff-cukier-parakilas-bryce.pdf. Dafoe, Allan. 2018. “AI Governance: A Research Agenda.” Oxford, UK: Future of Humanity Institute, Oxford University. www.fhi.ox.ac.uk/wp-content/uploads/GovAI-Agenda.pdf. Department for Science, Innovation & Technology. 2023. “Capabilities and risks from frontier AI: A discussion paper on the need for further research into AI risk.” October 25. London, UK: Government of the United Kingdom. www.gov.uk/government/publications/ frontier-ai-capabilities-and-risks-discussion-paper. Derechos Digitales. n.d. “A Latin American perspective on the use of AI systems by the State.” Santiago, Chile: Derechos Digitales. www.ohchr.org/sites/default/files/Documents/Issues/ DigitalAge/Submissions/CSOs/Derechos-Digitales.pdf. Edelman Trust Institute. 2024. 2024 Edelman Trust Barometer, Supplemental Report: Insights for the Tech Sector. NewYork, NY: Edelman Trust Institute. www.edelman.com/ sites/g/files/aatuss191/files/2024-03/2024%20 Edelman%20Trust%20Barometer%20Supplemental%20 Report%20Insights%20for%20Tech.pdf. Egan, Janet and Eric Rosenbach. 2023. “Biosecurity in the Age of AI: What’s the Risk?” Policy brief, November 6. Cambridge, MA: Belfer Center for Science and International Affairs. www.belfercenter.org/publication/ biosecurity-age-ai-whats-risk. 15Why Global South Countries Need to Care About Highly Capable AI El-Sayed, Seliem, Canfer Akbulut, Amanda McCroskery, Geoff Keeling, Zachary Kenton, Zaria Jalan, Nahema Marchal et al. 2024. “A Mechanism-Based Approach to Mitigating Harms from Persuasive Generative AI.” Preprint, arXiv, April 23. https://doi.org/10.48550/arXiv.2404.15058. Feldstein, Steven. 2019. “The Global Expansion of AI Surveillance.” Working Paper, September. Washington, DC: Carnegie Endowment for International Peace. https://carnegie-production-assets.s3.amazonaws.com/ static/files/files__WP-Feldstein-AISurveillance_final1.pdf. ———. 2021. The Rise of Digital Repression: How Technology is Reshaping Power, Politics, and Resistance. New York, NY: Oxford University Press. ———. 2024. “AI in war: Can advanced military technologies be tamed before it’s too late?” Bulletin of the Atomic Scientists, January 11. https://thebulletin.org/2024/01/ ai-in-war-can-advanced-military-technologies-betamed-before-its-too-late/#post-heading. Frantz, Erica. 2012. “How and Why do Dictatorships Survive? Lessons for the Middle East.” Bridgewater Review 31 (2): 16–18. https://vc.bridgew.edu/br_rev/vol31/iss2/7/. Frantz, Erica, Andrea Kendall-Taylor and Joseph Wright. 2020. “Digital Repression in Autocracies.” Users Working Paper Series 2020:27. Gothenburg, Sweden: V-Dem Institute. https://v-dem.net/media/publications/ digital-repression17mar.pdf. Freedom House. 2024. Freedom in the World 2024: The Mounting Damage of Flawed Elections and Armed Conflict. February. Washington, DC: Freedom House. https://freedomhouse.org/ sites/default/files/2024-02/FIW_2024_DigitalBooklet.pdf. Funk, Allie, Adrian Shahbaz and Kian Vesteinsson. 2023. “The Repressive Power of Artificial Intelligence.” In Freedom on the Net 2023, edited by Adrian Shahbaz, Allie Funk, Jennifer Brody, Kian Vesteinsson, Grant Baker, Cathryn Grothe, Matthew Barak, et al. Washington, DC: Freedom House. https://freedomhouse.org/sites/ default/files/2023-11/FOTN2023Final.pdf. Gabriel, Iason, Arianna Manzini, Geoff Keeling, Lisa Anne Hendricks, Verena Rieser, Hasan Iqbal, Nenad Tomašev et al. 2024. “The Ethics of Advanced AI Assistants.” Preprint, arXiv, April 28. https://doi.org/10.48550/arXiv.2404.16244. Gardizy, Anissa and Amir Efrati. 2024. “Microsoft and OpenAI Plot $100 Billion Stargate AI Supercomputer.” The Information, March 29. www.theinformation.com/articles/microsoftand-openai-plot-100-billion-stargate-ai-supercomputer. Gemini Team. 2023. “Gemini: A Family of Highly Capable Multimodal Models.” London, UK: Google DeepMind. https://assets.bwbx.io/documents/ users/iqjWHBFdfxIU/r7G7RrtT6rnM/v0. Georgieva, Kristalina. 2024. “AI Will Transform the Global Economy. Let’s Make Sure It Benefits Humanity.” January 14. www.imf.org/en/Blogs/Articles/2024/01/14/ ai-will-transform-the-global-economy-letsmake-sure-it-benefits-humanity. Gillani, Noor. 2023. “Will AI ever reach human-level intelligence? We asked five experts.” The Conversation, April 18. https://theconversation.com/will-ai-ever-reach-humanlevel-intelligence-we-asked-five-experts-202515. Gillespie, Nicole, Steve Lockey, Caitlin Curtis, Javad Pool and Ali Akbari. 2023. Trust in Artificial Intelligence: A Global Study. Queensland, AU: University of Queensland. https://assets.kpmg.com/content/dam/kpmg/au/ pdf/2023/trust-in-ai-global-insights-2023.pdf. Giri, Felipe Suárez. 2023. “Latin American Cities in the Age of AI: Navigating the Technological Revolution.” Harvard Kennedy School Student Policy Review, June 21. https://studentreview.hks.harvard.edu/latin-american-citiesin-the-age-of-ai-navigating-the-technological-revolution/. Harris, Jeremie, Edouard Harris and Mark Beall. 2023. “Survey of AI Technologies and AI R&D Trajectories.” Alexandria, VA: Gladstone AI. https://cdn.prod.website-files.com/ 62c4cf7322be8ea59c904399/ 65e83959fd414a488a4fa9a5_Gladstone%20 Survey%20of%20AI.pdf. Heath, Alex. 2024. “Mark Zuckerberg’s new goal is creating artificial general intelligence: And he doesn’t want to control it. Maybe.” The Verge, January 18. www.theverge.com/2024/1/18/24042354/ mark-zuckerberg-meta-agi-reorg-interview. Heikkilä, Melissa. 2024. “Google’s Astra is its first AI-foreverything agent.” MIT Technology Review, May 14. www.technologyreview.com/2024/05/14/1092407/ googles-astra-is-its-first-ai-for-everything-agent/. Heller, Patrick. 2022. “Democracy in the Global South.” Annual Review of Sociology 48: 463–84. https://doi.org/ 10.1146/annurev-soc-030320-123449. Hendrycks, Dan, Mantas Mazeika and Thomas Woodside. 2023. “An Overview of Catastrophic AI Risks.” Preprint, arXiv, October 9. https://doi.org/10.48550/arXiv.2306.12001. 16 CIGI Papers No. 311 — December 2024 • Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali Hiebert, Kyle. 2023. “Rising digital surveillance threatens Africa’s democratic progress.” ISS African Futures (blog), October 4. https://futures.issafrica.org/blog/2023/Rising-digitalsurveillance-threatens-Africas-democratic-progress. Hörster, Tim. 2021. “Utility of Artificial Intelligence to Authoritarian Governance.” Preprint, Social Science Research Network, September 15. https://papers.ssrn.com/sol3/ papers.cfm?abstract_id=4074964. Human Rights Watch. 2023. “DR Congo: Atrocities by RwandaBacked M23 Rebels.” News, February 6. New York, NY: Human Rights Watch. www.hrw.org/news/2023/02/06/ dr-congo-atrocities-rwanda-backed-m23-rebels. ICRC. 2019. Artificial intelligence and machine learning in armed conflict: A human-centred approach. Report, June 6. Geneva, Switzerland: ICRC. www.icrc.org/ en/document/artificial-intelligence-and-machinelearning-armed-conflict-human-centred-approach. Ipsos. 2022. “Global Opinions and Expectations about Artificial Intelligence: A Global Advisor survey.” January. Paris, France: Ipsos. www.ipsos.com/sites/ default/files/ct/news/documents/2022-01/Globalopinions-and-expectations-about-AI-2022.pdf. Jenkins, J. Craig and Augustine J. Kposowa. 1990. “Explaining Military Coups D’État: Black Africa, 1957–1984.” American Sociological Review 55 (6): 861–75. https://doi.org/10.2307/2095751. Jørgensen, Karen Helene. 2019. “The Meaning of Internet Access in Myanmar.” Master’s thesis, Norwegian University of Life Sciences. https://nmbu.brage.unit.no/nmbu-xmlui/ bitstream/handle/11250/2612476/ J%C3%B8rgensen2019.pdf?sequence=1. Juling, Dominik. 2023. “Future Bioterror and Biowarfare Threats for NATO’s Armed Forces until 2030.” Journal of Advanced Military Studies 14 (1): 118–43. https://doi.org/ 10.21140/mcuj.20231401005. Kapania, Shivani, Oliver Siy, Gabe Clapper, Azhagu Meena and Nithya Sambasivan. 2022. “‘Because AI is 100% right and safe’: User Attitudes and Sources of AI Authority in India.” Paper presented at the CHI Conference on Human Factors in Computing Systems. New Orleans, LA, April 29– May 5. https://doi.org/10.1145/3491102.3517533. Kaplan, Jared, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu and Dario Amodei. 2020. “Scaling Laws for Neural Language Models.” Preprint, arXiv, January 23. https://arxiv.org/pdf/2001.08361. Karnofsky, Holden. 2016. “Some Background on Our Views Regarding Advanced Artificial Intelligence.” Open Philanthropy (blog), May 6. www.openphilanthropy.org/ research/some-background-on-our-views-regardingadvanced-artificial-intelligence/#2-when-shouldwe-expect-transformative-artificial-intelligence. Kausch, Kristina, Aitor Bonsoms, Can Kasapoglu, Lewin Schmitt and Žilvinas Švedkauskas. 2022. Liberty’s Doom? Artificial Intelligence in Middle Eastern Security. European Institute of the Mediterranean. www.euromesco.net/ wp-content/uploads/2022/07/Policy-Study27.pdf. Kendall-Taylor, Andrea, Erica Frantz and Joseph Wright. 2020. “The Digital Dictators: How Technology Strengthens Autocracy.” Foreign Affairs, February 6. www.foreignaffairs.com/ articles/china/2020-02-06/digital-dictators. Kenton, Zachary, Ramana Kumar, Sebastian Farquhar, Jonathan Richens, Matt MacDermott and Tom Everitt. 2022. “Discovering Agents.” Preprint, arXiv, August 24. https://doi.org/10.48550/arXiv.2208.08345. Klark, Arthur and Gabriela Sagun. 2023. “Misinformation, Hate Speech and Ethno-Religious Tensions in Myanmar.” Analysis, April 27. Washington, DC: United States Institute of Peace. www.usip.org/publications/2023/04/misinformationhate-speech-and-ethno-religious-tensions-myanmar. Knuutila, Aleksi, Lisa-Maria Neudert and Philip N. Howard. 2022. “Who is afraid of fake news? Modeling risk perceptions of misinformation in 142 countries.” Harvard Kennedy School Misinformation Review 3 (3): 1–13. https://doi.org/10.37016/mr-2020-97. Korinek, Anton and Megan Juelfs. 2022. “Preparing for the (non-existent?) future of work.” Center on Regulation and Markets Working Paper 3. Washington, DC: Brookings Institution. www.brookings.edu/articles/ preparing-for-the-non-existent-future-of-work/. Korinek, Anton and Joseph E. Stiglitz. 2021. “Artificial Intelligence, Globalization, and Strategies for Economic Development.” National Bureau of Economic Research Working Paper 28453. August. www.nber.org/system/ files/working_papers/w28453/w28453.pdf. Krakovna, Victoria, Jonathan Uesato, Vladimir Mikulik, Matthew Rahtz, Tom Everitt, Ramana Kumar, Zac Kenton, Jan Leike and Shane Legg. 2020. “Specification gaming: the flip side of AI ingenuity.” Medium, April 21. https://deepmindsafetyresearch.medium.com/specificationgaming-the-flip-side-of-ai-ingenuity-c85bdb0deeb4. 17Why Global South Countries Need to Care About Highly Capable AI Lehdonvirta, Vili, Bóxī Wú and Zoe Hawkins. 2024. “Compute North vs. Compute South: The Uneven Possibilities of Compute-based AI Governance Around the Globe.” Preprint, SocArXiv Papers, August 25. https://doi.org/10.31235/osf.io/8yp7z. Lehmann, Jean-Pierre. 1985. “Dictatorship and development in Pacific Asia: wider implications.” International Affairs 61(4): 591–606. https://doi.org/10.2307/2617705. Liu, Junfeng, Christopher Symons and Ranga Raju Vatsavai. 2022. “Persona-Based Conversational AI: State of the Art and Challenges.” Preprint, arXiv, December 4. https://doi.org/10.48550/arXiv.2212.03699. Maas, Matthijs M., Kayla Lucero-Matteucci and Di Cooke. 2023. “Military Artificial Intelligence as a Contributor to Global Catastrophic Risk.” In The Era of Global Risk: An Introduction to Existential Risk Studies, edited by S. J. Beard, Martin Rees, Catherine Richards and Clarissa Rios Rojas, 237–84. Cambridge, UK: Open Book Publishers. https://books.openbookpublishers.com/ 10.11647/obp.0336.10.pdf. Mandavia, Megha. 2024. “AI Is Coming for India’s Famous Tech Hub.” The Wall Street Journal, August 6. www.wsj.com/tech/ ai/ai-is-coming-for-indias-famous-tech-hub-97261085. Manvi, Rohin, Samar Khanna, Marshall Burke, David Lobell and Stefano Ermon. 2024. “Large Language Models are Geographically Biased.” Preprint, arXiv, October 5. https://doi.org/10.48550/arXiv.2402.02680. Matlosa, Khabele. 2023. “Global trends and impact of democratic recession: Hard choices for the Global South.” South African Journal of International Affairs 30 (3): 337–55. https://doi.org/10.1080/10220461.2023.2269149. Mora, Frank and Brian Fonseca. 2019. “It’s Not the 1970s Again for Latin America’s Militaries. Here’s Why.” Americas Quarterly, December 11. www.americasquarterly.org/article/its-notthe-1970s-again-for-latin-americas-militaries-heres-why/. Murgia, Madhumita. 2023. “OpenAI chief seeks new Microsoft funds to build ‘superintelligence.’” Financial Times, November 13. www.ft.com/content/ dd9ba2f6-f509-42f0-8e97-4271c7b84ded. Ngo, Richard. 2020. “AGI Safety From First Principles.” AI Alignment Forum, September 28. www.alignmentforum.org/s/ mzgtmmTKKn5MuCzFJ. ———. 2023. “Visualizing the deep learning revolution.” Medium, January 5. https://medium.com/@richardcngo/ visualizing-the-deep-learning-revolution-722098eb9c5. Ngo, Richard, Lawrence Chan and Sören Mindermann. 2024. “The Alignment Problem from a Deep Learning Perspective.” Preprint, arXiv, March 19. https://arxiv.org/pdf/2209.00626. O’Brien, John T. and Cassidy Nelson. 2020. “Assessing the Risks Posed by the Convergence of Artificial Intelligence and Biotechnology.” Health Security 18 (3): 219–27. https://liebertpub.com/doi/pdf/10.1089/hs.2019.0122. Omeje, Kenneth. 2018. “Extractive Economies and Conflicts in the Global South: Re-Engaging Rentier Theory and Politics.” In Extractive Economies and Conflicts in the Global South: Multi-Regional Perspectives on Rentier Politics, edited by Kenneth Omeje, 1–25. Abingdon, UK: Routledge. Onome. 2024. “How AI is Keeping Old Languages Alive.” AutoGPT by Mindstream, June 26. https://autogpt.net/ the-impact-of-ai-in-languages-preservation/#:~:text=AI%20 aids%20language%20documentation%20and,and%20 supporting%20real%2Dtime%20translation. Onyekwena, Chukwuka and Mma Amara Ekeruche. 2019. “The Global South and development assistance.” Commentary, September 16. Washington, DC: Brookings Institution. www.brookings.edu/articles/ the-global-south-and-development-assistance/. Ord, Toby. 2020. The Precipice: Existential Risk and the Future of Humanity. New York, NY: Hachette Books. Palik, Júlia, Anna M. Obermeier and Siri A. Rustad. 2022. “Conflict Trends in Africa, 1989–2021.” PRIO Paper. Oslo, Norway: Peace Research Institute Oslo. www.prio.org/publications/13215. Patel, Nilay. 2023. “Inside Google’s big AI shuffle — and how it plans to stay competitive, with Google DeepMind CEO Demis Hassabis.” The Verge, July 10. www.theverge.com/23778745/ demis-hassabis-google-deepmind-ai-alphafold-risks. Phyo. 2023. “The Digital Landscape of Activism in Myanmar.” EngageMedia, November 29. https://engagemedia.org/ 2023/youth-myanmar/. Piper, Kelsey. 2020. “The case for taking AI seriously as a threat to humanity: Why some people fear AI, explained.” Vox, October 15. www.vox.com/future-perfect/2018/ 12/21/18126576/ai-artificial-intelligencemachine-learning-safety-alignment. Prabhakaran, Vinodkumar, Rida Qadri and Ben Hutchinson. 2022. “Cultural Incongruencies in Artificial Intelligence.” Preprint, arXiv, November 19. https://doi.org/10.48550/ arXiv.2211.13069. 18 CIGI Papers No. 311 — December 2024 • Cecil Abungu, Marie Victoire Iradukunda, Duncan Cass‑Beggs, Aquila Hassan and Raqda Sayidali Privitera, Daniel, Tamay Besiroglu, Rishi Bommasani, Stephen Casper, Yejin Choi, Danielle Goldfarb, Hoda Heidari and Leila Khalatbari. 2024. International Scientific Report on the Safety of Advanced AI: Interim Report. May 17. London, UK: Department for Science, Innovation and Technology. www.gov.uk/government/publications/internationalscientific-report-on-the-safety-of-advanced-ai. Putzier, Konrad. 2023. “Next Wave of Remote Work Is About Outsourcing Jobs Overseas.” The Wall Street Journal, April 11. www.wsj.com/articles/next-wave-of-remotework-is-about-outsourcing-jobs-overseas-54af39ba. Ritchie, Hannah, Veronika Samborska, Natasha Ahuja, Esteban Ortiz-Ospina and Max Roser. 2023. “Global Education.” OurWorldinData.org. https://ourworldindata.org/globaleducation?insight=children-across-the-world-receive-verydifferent-amounts-of-quality-learning#key-insights. Roffey, Roger. 2004. “Biological weapons and potential indicators of offensive biological weapon activities.” In SIPRI Yearbook 2004: Armaments, Disarmament and International Security, edited by Stockholm International Peace Research Institute, 557–71. Oxford, UK: Oxford University Press. www.sipri.org/yearbook/2004/13. Roser, Max. 2023. “Technology over the long run: zoom out to see how dramatically the world can change within a lifetime.” OurWorldinData.org, February 22. https://ourworldindata.org/technology-long-run#. Rustad, Siri A. 2024. “Conflict Trends: A Global Overview, 1946–2023.” PRIO Paper. Oslo, Norway: Peace Research Institute Oslo. www.prio.org/publications/14006. Schapera, Isaac. 1955. A Handbook of Tswana Law and Custom. 2nd ed. London, UK: Frank Cass & Co. Schmidt, Eric. 2023. “Innovation Power: Why Technology Will Define the Future of Geopolitics.” Foreign Affairs, February 28. www.foreignaffairs.com/united-states/ eric-schmidt-innovation-power-technology-geopolitics. Shavit, Yonadav, Sandhini Agarwal, Miles Brundage, Steven Adler, Cullen O’Keefe, Rosie Campbell, Teddy Lee et al. 2023. “Practices for Governing Agentic AI Systems.” OpenAI paper, December 14. https://cdn.openai.com/ papers/practices-for-governing-agentic-ai-systems.pdf. Siani, Alessandro and Imogen Green. 2023. “Scientific Misinformation and Mistrust of COVID-19 Preventive Measures among the UK Population: A Pilot Study.” Vaccines1(2):301. https://doi.org/10.3390/vaccines11020301. Sotala, Kaj and Roman V. Yampolskiy. 2015. “Responses to catastrophic AGI risk: a survey.” Physica Scripta 90: 018001. https://iopscience.iop.org/ article/10.1088/0031-8949/90/1/018001/pdf. Stoelker, Tom. 2018. “American TV, International Perceptions, and Reality.” Fordham Now, March 28. https://now.fordham.edu/ inside-fordham/faculty-reads/american-tv-internationalperceptions-and-reality/. Szczepański, Marcin. 2019. “Economic impacts of artificial intelligence (AI).” Briefing, July. Brussels, Belgium: European Parliamentary Research Service. www.europarl.europa.eu/ RegData/etudes/BRIE/2019/637967/ EPRS_BRI(2019)637967_EN.pdf. Taleb, Abdelrahman Abu. 2024. “Evidence of Iran and UAE drones used in Sudan war.” BBC, June 12. www.bbc.com/ news/articles/c2vvjz652j1o. Tan, Jun-E, Kathleen Azali and Katerina Francisco. 2021. Governance of Artificial Intelligence (AI) in Southeast Asia. Melbourne, AU: EngageMedia. December. https://engagemedia.org/ wp-content/uploads/2021/12/Engage_Report-Governanceof-Artificial-Intelligence-AI-in-Southeast-Asia_12202021.pdf. Tao, Yan, Olga Viberg, Ryan S. Baker and René F. Kizilcec. 2024. “Cultural Bias and Cultural Alignment of Large Language Models.” Preprint, arXiv, June 26. https://arxiv.org/abs/2311.14096v2. Tech Wire Asia. 2018. “The impact of AI is as great as the steam engine: McKinsey.” Tech Wire Asia, August 31. https://techwireasia.com/09/2018/ai-will-bringglobal-growth-on-par-with-steam-engine/. The Economist. 2023. “ChatGPT could replace telemarketers, teachers and traders.” The Economist, April 14. www.economist.com/graphic-detail/2023/04/14/ chatgpt-could-replace-telemarketers-teachers-and-traders. Timcke, Scott and Hanani Hlomani. 2024. Decoding the Ballot: How Might AI Reshape Democracy on the African Continent? Cape Town, South Africa: Research ICT Africa. https://researchictafrica.net/research/how-might-aireshape-democracy-on-the-african-continent/. Todaro, Michael P. and Stephen C. Smith. 2020. Economic Development. 13th ed. Harlow, UK: Pearson. Transparency International. 2024. Corruption Perceptions Index 2023. Berlin, Germany: Transparency International. https://transparencia.org.es/wp-content/uploads/ Report_CPI2023_English.pdf.